Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Interpretation
Claim 1 is a process claim, The claim recites the following contingent limitation: “if it is determined that at least one working set of the plurality of working sets requires memory usage above the pre-determined threshold, identifying a working set of the plurality of working sets which requires memory usage above the pre-determined threshold”. This limitation is contingent because it only occurs “if it is determined that at least one working set of the plurality of working sets requires memory usage above the pre-determined threshold”. Due to this contingent limitation, the BRI of the claim only requires the following limitations: “for each of the plurality of layers, determining a tensor working set, wherein the tensor working set comprises that consume memory with respect to the respective layer” and “determining whether at least one working set of the plurality sets requires memory usage above the pre-determined threshold” for completeness of the Office Action even though the limitations “if it is determined that at least one working set of the plurality of working sets which requires memory usage above the pre-determined threshold; identifying at least one layer responsible for the memory usage above the pre-determined threshold in the identified working set; and pruning the identified at least one layer” in claim 1 are not required to be rejected per below.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 9, 10, and 13 rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 9, 10, and 13, the phrase "preferably" renders the claim indefinite because it is unclear whether the limitation(s) following the phrase are part of the claimed invention. See MPEP § 2173.05(d).
Regarding Claim 3 the term “highest” in claim 3 is a relative term which renders the claim indefinite. The term “highest” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. what is deemed highest amount of memory for the working at has been rendered indefinite.
Regarding Claim 7 the term “maximum” in claim 7 is a relative term which renders the claim indefinite. The term “maximum” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. what is deemed maximum memory usage for the intermediate representation.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim’s 1-15 are rejected as being directed to an abstract idea without significantly more.
Step 2A prong 1: Keeping the above claim interpretation in mind, looking at all the claim limitations of claim 1 at the claim recites the abstract idea of “for each of the plurality of layers, determining a tensor working set, wherein the tensor working set comprises tensors that consume memory with respect to the respective layer” (Determining a tensor consuming memory is a mental process that can be done with the aid of pen and paper) , ”determining whether at least one working set of the plurality of working sets requires memory usage above a pre-determined threshold” (Determining and identifying a working set of tensors consuming memory above a threshold is a mental process aid of pen and paper) , “if it is determined that at least one working set of the plurality of working sets requires memory usage above the pre-determined threshold, identifying a working set of the plurality of working sets which requires memory usage above the pre-determined threshold” (Identifying working sets that requires memory usage above a threshold is a mental process that can be done with the aid of pen and paper), “identifying at least one layer responsible for the memory usage above the pre-determined threshold in the identified working set” (The process of identifying a layer responsible for memory usage above a threshold in the working set is a mental process that can be done with the aid of pen and paper).
Step 2A prong 2: Note that due to the contingent limitations noted above, the claim only requires the first to limitations of the claim. Therefore, given the BRI of the claim, the claim does not require the limitation of “pruning the identified at least one layer”. However, if it did, this pruning step would only amount to insignificant extra-solution activity (MPEPE 2106.05(g)) and “apply it” as it fails to recite any additional elements beyond just pruning to take this limitation beyond just applying technique to be done on the tensor working set (MPEO 2106.05(f)). Furthermore, the claim only additionally recites that it is “computer-implemented” which only amounts to “applying” the abstract idea with use of a generic computer component (MPEP 2106.05(f)). None of these additional limitations, taken either alone or in combination, amount to a practical application.
Step 2B again keeping the claim interpretation above in mind, the claim only requires the first two limitations of the claim. Therefore, given the BRI of the claim, the claim does not require the limitation of “pruning the identified at least one layer”. However, if it did, this pruning step would only amount to well understood, routine and conventional activity (see specification, page 1, lines 21-23 discuss that pruning is, “One typical strategy to minimize the requirements of DNNs…”, additionally, page 1, line 29-page 2, line 28 cite papers that discuss various pruning techniques known in the art). Furthermore, the claim only additionally recites that it is “computer-implemented” which only amounts to “applying” the abstract idea with use of a generic computer components (MPEP 2106.05(f)). None of these additional limitations, taken either alone or in combination with the other limitations of the claim, amount to significantly more than the abstract idea itself.
Regarding Claim 2
Step 1: A process as above.
Step 2A Prong 1: The claim recites “where in each steps of identifying and pruning are repeated until every working set of the plurality of working sets requires memory below the pre-determined threshold”, this limitation recites the abstract idea claim 1
Step 2A Prong 2 & Step 2B: There are no additional elements recited so the claim does
not provide a practical application and is not considered to be significantly more. As such,
the claim is patent ineligible.
Regarding Claim 3
Step 1: A process as above
Step 2A Prong 1: The claim recites:
“Where in each step of identifying, the working set which requires a highest amount of memory, identified” These limitation(s) encompass a mental process with the aid of pen and paper of using observation to identify the working set that requires the highest amount of memory
Step 2A Prong 2 & Step 2B: There are no additional elements recited so the claim does
not provide a practical application and is not considered to be significantly more. As such,
the claim is patent ineligible.
Regarding Claim 4
Step 1: A process as above
Step 2A Prong 1: The claim recites the abstract idea claim 1
Step 2A Prong 2: The additional limitation of “pruning the at least one identified layer comprises reducing a number of channels of the at least one identified layer” amounts to nothing more than “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of training a machine learning model with previously determined data
Step 2B: The pruning channels limitation is a technique well understood, routine and conventional (see specification, page 1, lines 21-23 discuss that pruning is, “One typical strategy to minimize the requirements of DNNs…”, additionally, page 1, line 29-page 2, line 28 cite papers that discuss various pruning techniques known in the art) and the additional limitation.
Regarding Claim 5
Step 1: A process as above
Step 2A Prong 1: The claim recites the abstract idea claim 1
Step 2A Prong 2: The additional limitation of “wherein pruning the at least one identified layer comprises removing the at least one identified layer” amounts to nothing more than “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of training a machine learning model with previously determined data
Step 2B: The pruning layers limitation is a technique well understood, routine and conventional (see specification, page 1, lines 21-23 discuss that pruning is, “One typical strategy to minimize the requirements of DNNs…”, additionally, page 1, line 29-page 2, line 28 cite papers that discuss various pruning techniques known in the art) and the additional limitation.
Regarding Claim 6
Step 1: A process as above.
Step 2A Prong 1: The claim recites the abstract limitation of “the working sets which requires maximum memory usage is determined based on an architecture of the artificial neural network graph”. Determining memory usage with the aid of a graph is a mental process that a human can perform. The additional limitations do not integrate the mental process into practical application or add significantly more to the mental process.
Step 2A Prong 2: The additional limitations of “the working sets which requires maximum memory usage is determined based on an architecture of the artificial neural network graph” amounts to nothing more than mere data gathering using the neural network graph to represent memory usage
Step 2B: using an artificial neural network graph to represent memory usage is something that is well understood, routine, and conventual in the art
Regarding Claim 7
Step 1: A process as above.
Step 2A Prong 1: The claim recites the abstract limitation of “determining intermediate representation of the artificial neural network graph; wherein the working set of the plurality of working sets which requires maximum memory usage is determined based on the intermediate representation.” This claim encompasses a mental process where a human can observe the intermediate representation and determine what set is consuming the maximum memory usage.
Step 2A Prong 2 & Step 2B: There are no additional elements recited so the claim does
not provide a practical application and is not considered to be significantly more. As such,
the claim is patent ineligible.
Regarding Claim 8
Step 1: A process as above.
Step 2A Prong 1: The claim recites the ideas of claim 1
Step 2A Prong 2: The additional limitations of “wherein once every working set of the plurality of working sets requires memory below the pre-determined threshold, the artificial neural network graph after pruning is retrained from scratch or fine-tuned from a previous training” amounts to nothing more than mere instructions to apply an exception with the use of pruning, and training, and finetuning just being used to train the already established abstract idea (2106.05(f)).
Step 2B: The pruning layers limitation is a technique well understood, routine and conventional (see specification, page 1, lines 21-23 discuss that pruning is, “One typical strategy to minimize the requirements of DNNs…”, additionally, page 1, line 29-page 2, line 28 cite papers that discuss various pruning techniques known in the art) and the additional limitation. The process of training and fine-tuning is a process that is a process that is well understood, routine, and conventional in the art.
Regarding Claim 9
Step 1: A process as above.
Step 2A Prong 1: The claim recites the abstract ideas of claim 1.
Step 2A Prong 2: The additional limitation of “wherein the at least one identified layer is pruned based on an importance metric, wherein preferably the importance metric is provided by user input” amounts to nothing more than “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: high level recitation of training a machine learning model with previously determined data.
Step 2B: The additional limitation of pruning with the use of a metric is a technique that is well understood, routine and conventional (conventional (see specification, page 1, lines 21-23 discuss that pruning is, “One typical strategy to minimize the requirements of DNNs…”, additionally, page 1, line 29-page 2, line 28 cite papers that discuss various pruning techniques known in the art).
Regarding Claim 10
Step 1: A process as above.
Step 2A Prong 1: The limitation “the importance metric is evaluated based on an importance metric” is a mental process that can be done with the aid of pen and paper. A human can come up with their own metric and use it as an importance metric.
Step 2A Prong 2 & Step 2B: There are no additional elements recited so the claim does
not provide a practical application and is not considered to be significantly more. As such,
the claim is patent ineligible.
Regarding Claim 11
Step 1: A process as above.
Step 2A Prong 1: The claim recites “generating a report comprising at least one of a layer summary report, a tensor summary report, or a working set summary report”. Generating a report consisting of either a layer summary report, a tensor summary report, or a working set summary report here is a mental process that can be done with the aid of pen and paper.
Step 2A Prong 2: The additional limitation of “generating a report comprising at least one of a layer summary report, a tensor summary report, or a working set summary report” is a process that doesn’t add anything significantly more to the abstract idea beyond just applying a generic computer method to the abstract idea (2106.05(f)).
Step 2B: The additional elements, taken either alone or in combination with other limitations of the claim, do not amount to significantly more than the abstract idea itself. Generating a report for an abstract idea does not significantly more to the abstract idea to make it an inventive concept.
Regarding Claim 12
Step 1: A process as above.
Step 2A Prong 1: The claim recites the ideas of claim 1.
Step 2A Prong 2: the additional limitation “the artificial neural network and or the predetermined threshold are provided by user input” do not add anything significantly more to the claim to stop it from being abstract. A user input for the artificial neural network and or threshold is just a form of selecting a particular data source to be used for the abstract idea which is not enough to integrate the abstract idea into a practical application
Step 2B: The additional elements, taken alone or in combination, don not represent significantly more than the abstract idea itself. Using a user input to represent the artificial neural network or threshold does not add enough to the abstract idea to make it an inventive concept.
Regarding Claim 13
Step 1: A process as above.
Step 2A Prong 1: The claim recites the ideas of claim 1.
Step 2A Prong 2: the additional limitation “wherein the artificial neural network graph is to be deployed on a resource-constrained embedded system after pruning” does not add anything significantly more to the abstract idea beyond just applying a generic computer equipment to execute the abstract idea (2106.05(f)). The additional limitation “wherein preferably the embedded system is a mobile computing device, a mobile phone, a tablet computing device, an automotive compute platform, or an edge device” does not add anything significantly more to the abstract idea beyond just applying a generic computer equipment to execute the abstract idea (2106.05(f)).
Step 2B: The additional elements, taken alone or in combination, don not represent significantly more than the abstract idea itself. Using an embedded system to deploy the neural network graph after pruning does not add enough to the abstract idea to take it to being an inventive concept.
Regarding Claim 14:
Step 1: A process as above.
Step 2A Prong 1: See the analysis of Claim 1.
Step 2A Prong 2 & Step 2B: There are no additional elements recited so the claim does
not provide a practical application and is not considered to be significantly more. As such,
the claim is patent ineligible.
Regarding Claim 15:
Step 1: A process as above.
Step 2A Prong 1: See the analysis of Claim 1.
Step 2A Prong 2 & Step 2B: There are no additional elements recited so the claim does
not provide a practical application and is not considered to be significantly more. As such,
the claim is patent ineligible.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
(a) the invention was known or used by others in this country, or patented or described in a printed publication in this or a foreign country, before the invention thereof by the applicant for a patent.
Regarding claims 1, 14 and 15 are rejected under 102 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”)
Mateev teachs A computer implemented method for optimizing memory usage of an artificial neural network graph comprising a plurality of layers ([Abstract] mentions the Neural network may contain a plurality of layers ) and a plurality of tensors (Abstract mentions the use of tensors columns which is being interpreted as a plurality of tensors) the method comprising the steps: for each of the plurality of layers, determining a tensor working set, wherein the tensor working set comprises tensors that consume memory with respect to the respective layer ([Abstract] teaches the relationship between kernels and layers where one layer includes one or more kernels. It is being interpreted off of what is described in the patent application [160], [0027] that there is a relationship between kernels and tensors where the presence of kernels implies the presence of tensors. The kernels in the patent application are being pruned to fit below a cache memory threshold so with that in mid this limitation teaches tensors that consume memory with respect to the respective layer.), determining whether at least one working set of the plurality of working sets requires memory usage above a pre-determined threshold ([0102] teaches kernels computation not exceeded a cache memory threshold which functions as a pre-determined threshold), if it is determined that at least one working set of the plurality of working sets which requires memory usage above the pre-determined threshold, identifiying a working set of the plurality of working sets which requires memory usage above the pre-determined threshold ([0161] Teaches a conditional statement where when the kernels are large they would be pruned to fit below the memory threshold); identifying at least one layer responsible for the memory usage above the pre-determined threshold in the working set; and pruning the identified at least one layer ([0161] teaches the limitation of identifying layers with the identification of kernels above the memory threshold and those kernels go through sparsification which is another word for pruning.
Regards Claims 14 and 15, they are rejected under Claim 1.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 2, 3 and 5 are rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Ji et al (US 20180232640) (“Ji”).
Regarding Claim 2 Mateev teaches all the limitations of Claim 1.
Mateev does not teach wherein the steps of identifying and pruning are repeated until every working set of the plurality of working sets requires memory below the pre-determined threshold.
However, Ji teaches wherein the steps of identifying and pruning are repeated until every working set of the plurality of working sets requires memory below the pre-determined threshold ([0051] teaches iteration-based pruning below a threshold).
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the repeated pruning technique of Ji. Doing so would ensure lower computation loads and lower power consumption ([0003] “Deep learning architectures, especially convolutional deep neural networks have been used in artificial intelligence (AI) and computer vision fields. These architectures have been shown to produce results on tasks including visual object recognition, detection and segmentation. However, such architectures may have a large number of parameters, resulting high computational loads and increased power consumption.”
Regarding Claim 3 Mateev teaches all limitations of Claim 1.
Mateev also teaches wherein in each step of identifying, the working set which requires a highest amount of memory is identified ([0102] teaches this limitation with the aim of the kernels not exceeding a predefined memory threshold).
Regarding Claim 5 Mateev teaches all the limitations of Claim 1.
Mateev does not teach the limitation wherein pruning the at least one identified layer comprises removing the at least one identified layer.
However, Ji teaches wherein pruning the at least one identified layer comprises removing the at least one identified layer ([Abstract] teaches pruning a layer in a neural network. The pruning process have you remove components of a neural network so pruning a layer is being interpreted as removing the layer.
Mateev and Ji are analogous are because they relate to neural networks and pruning.
Regarding Claim 4, it is rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view Ye et al (NPL: Channel Pruning via Optimal Thresholding) (“Ye”).
Mateev teaches all the limitations of Claim 1.
Mateev does not teach wherein each layer comprises a respective plurality of channels; and wherein pruning the at least one identified layer comprises reducing a number of channels of the at least one identified layer.
However, Ye does teach wherein each layer comprises a respective plurality of channels ([Abstract] teaches the limitation of a plurality of channels); and wherein pruning the at least one identified layer comprises reducing a number of channels of the at least one identified layer ([Abstract] teaches the limitation of pruning channels associated with layers).
Mateev and Ye are analogous are because they relate to neural networks and pruning.
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the channel pruning of Ye. Doing so would allow for more optimal pruning depending on the different layers ([Abstract] “The predefined global threshold based designs ignore the variation among different layers and weights distribution, therefore, they may often result in sub-optimal performance caused by over pruning or under-pruning. In this paper, we present a simple yet effective method, termed Optimal Thresholding (OT), to prune channels with layer dependent thresholds that optimally separate important from negligible channels. By using OT, most negligible or unimportant channels are pruned to achieve high sparsity while minimizing performance degradation.”).
Claims 6 and 7, they are rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Fu et al (NPL: Automatic Generation of High-Performance Inference Kernels for Graph Neural Networks on Multi-Core Systems) (“Fu”).
Regarding claim 6, Mateev teaches all the limitations of Claim 1.
Mateev does not teach determining an intermediate representation of the artificial neural network graph; wherein the working set of the plurality of working sets which requires maximum memory usage is determined based on an architecture of the artificial neural network graph.
However, Fu teaches wherein the working set of the plurality of working sets which requires maximum memory usage is determined based on an architecture of the artificial neural network graph ([4.2 Memory Usage] teaches measuring maximum memory usage of a neural network graph).
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the GNN representation from Fu for memory usage. Doing so would improve GNN memory efficiency (4.2 “The result shows Gin is able to save a large amount of memory for different input sizes. Especially for large graphs, Gin is able to achieve an average reduction of 86% compared to DGL, 72% to Tensorflow and 92% to Pytorch-geometric.”).
Regarding Claim 7 Mateev teaches all the limitations of Claim 1
Mateev does not teach wherein the working set of the plurality of working sets which requires maximum memory usage is determined based on an architecture of the artificial neural network graph
However, Fu teaches determining an intermediate representation of the artificial neural network graph ([Abstract] teaches an intermediate representation being an artificial neural network graph); wherein the working set of the plurality of working sets which requires maximum memory usage is determined based on the intermediate representation ([4.2] teaches using an intermediate representation to measure maximum memory usage)
Mateev and Fu are analogous art because they relate to the use of neural networks
Regarding claim 8, it is rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Zhou et al (NPL: Accelerating Large Scale Real-Time GNN Inference using Channel Pruning (“Zhou”), and Zhang et al (US 20210397965) (“Zhang”).
Mateev teaches all the limitations of Claim 1. Mateev also teaches wherein once every working set of the plurality of working sets requires memory below the pre-determined threshold ([0161] Mateev teaches pruning at least one kernel below a memory threshold this could reasonably be interpreted to accommodate all kernels which would address this limitation)
Mateev does not teach the artificial neural network graph after pruning is re-trained from scratch or fine-tuned from a previous training
However, Zhou teaches the artificial neural network graph after pruning is re-trained ([4.1] Zhou teaches retraining of the pruned methods and the pruning is being executed on a GNN which would address this limitation).
It would have been It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the model retraining of Zhou. Doing so would save memory ([Abstract] “We demonstrate that the pruned GNN models greatly reduce computation and memory usage with little accuracy loss.”).
Zhang teaches or fine-tuned from a previous training ([0094] Zhang teaches fine tuning a neural network after pruning and the use of graph diffusion implies the presence of a neural network graph to address this limitation)
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the graph diffusion-based pruning of Zhang. Doing so would allow for more important parameters to get pruned ([Abstract] “estimate an importance of parameters of a neural network based on a graph diffusion process over at least one layer of the neural network; determine the parameters of the neural network that are suitable for pruning or sparsification; remove neurons of the neural network to prune or sparsify the neural network”).
Mateev, Zhou and Zhang are analogous are because they relate to neural networks and pruning
Regarding Claim 9 it is rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Zhang et al (US 20210397965) (“Zhang”).
Mateev does not teach wherein the at least one identified layer is pruned based on an importance metric, wherein preferably the importance metric is provided by user input
However, Zhang does teach wherein the at least one identified layer is pruned based on an importance metric ([0074] Teaches a technique that identifies which neuron to prune, this technique functions as an importance metric and it is being used to identify a pruning target).
Mateev and Zhang are analogous are because they relate to neural networks and pruning.
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the graph diffusion-based pruning of Zhang. Doing so would allow for more important parameters to get pruned ([Abstract] “estimate an importance of parameters of a neural network based on a graph diffusion process over at least one layer of the neural network; determine the parameters of the neural network that are suitable for pruning or sparsification; remove neurons of the neural network to prune or sparsify the neural network”).
Regarding Claim 10 it is rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Zhang et al (US 20210397965) (“Zhang”) and LI et al (US 20230084203 A1) (“Li”).
Mateev and Zhang teaches all the limitations of Claim 9.
Mateev does not teach the computer implemented method preferably further comprising the following step: training the artificial neural network graph before evaluating the importance metric
However, LI teaches the computer implemented method preferably further comprising the following step: training the artificial neural network graph before evaluating the importance metric ([0081] Li teaches a methodology with pruning with a neural network graph that includes a training process. It ends with an architecture with the highest evaluation accuracy being selected and that can function as a importance metric).
Mateev Zhang, and Li are analogous are because they relate to neural networks and pruning.
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the graph diffusion-based pruning of Zhang and the artificial neural network graph training of Li. Doing so would result a more efficient neural network ([0005] “Accordingly, what is needed are systems and methods for channel pruning in neural networks for improved efficiency and performance.”).
Regarding Claim 11, it is rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Labarge et al (NPL: Neural Network Pruning For ECG Arrhythmia Classification).
Mateev teaches all the limitations of Claim 1.
Mateev does not teach generating a report comprising at least one ofa layer summary report, a tensor summary report, or a working set summary report.
However, Labarge teaches generating a report comprising at least one of a layer summary report, a tensor summary report, or a working set summary report (Figure 3.6 shows a model summary report showcasing the layers after pruning, this is being interpreted as covering the three different reports mentioned for this limitation).
Matev, Zhang and Labarge are analogous art because they relate to neural networks and pruning.
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the graph diffusion-based pruning of Zhang, and the summary report generation of Labarge. Doing so would allow for more accurate pruning with model finetuning “page 116 “Results for improving finetuning with the Taylor-expansion pruning method were marginally successful. While finetuning could be accelerated by a factor of 2× using the mixed 1:4 ratio locking scheme, it results in a 5.52% drop in accuracy for NAVS ECG and a 3.72% accuracy drop for CIFAR-10.”).
Regarding Claim 12, it is rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Zhang et al (US 20210397965) and Lou et al (US 12488249) (“Lou”)
Mateev teaches all the limitations of Claim 1.
Mateev does not teach wherein the artificial neural network or the pre-determined threshold are provided by user input.
However, Lou teaches wherein the artificial neural network ([Summary] teaches a prediction setting which is based on a user input and that setting is used to modify the neural network, with BRI in mind this is being interpreted as an artificial neural network provided by user input) and or the pre-determined threshold are provided by user input (Claim 10 teaches an accuracy threshold based on a user input).
Mateev and Lou are analogous art because they relate to neural networks and pruning.
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the artificial neural network and threshold user input method of Lou. Doing so would create a more secure neural network (Background/Summary (1) “The disclosure relates to a system and method for providing secure neural network inference on encrypted data without decryption..”).
Regarding Claim 13 it is rejected under 103 as being unpatentable over Mateev et al (US 20200160182) (“Mateev”), in view of Hao et al (CN111260049A) (“Hao”).
Mateev teaches all the limitations of Claim 1.
Mateev does not teach wherein the artificial neural network graph is to be deployed on a resource-constrained embedded system after pruning; wherein preferably the embedded system is a computing device, a mobile phone, a table computing device, an automotive compute platform or an edge device.
However, Hao does teach wherein the artificial neural network graph is to be deployed on a resource-constrained embedded system after pruning; wherein preferably the embedded system is a computing device, a mobile phone, a table computing device, an automotive compute platform or an edge device ([Detailed Description] Hao teaches a pruned neural network model being put on a domestic embedded system. The pruned neural network model is being interpreted as a neural network graph and the use of a domestic embedded system here would cover all the mentioned embedded system devices).
Mateev and Hao are analogous art because they relate to neural networks and pruning.
It would have been obvious to a person skilled in the art before the effective filling date of the claimed invention to combine Mateev with the deployment of the pruned neural network model to a domestic embedded device. Doing so would allow for neural network models to be deployed on different devices without the need to be networked all the time (Background “The traditional convolutional neural network has more network parameters, is limited by huge storage occupation and calculation cost of a convolutional neural network model, and an embedded system with limited resources is difficult to directly operate the convolutional neural network with high requirements. At present, most of the application of the neural network is a solution of 'local + cloud', the solution is undoubtedly time-consuming, devices need to be networked all the time, and certain requirements are also made on network speed and network bandwidth.”).
Conclusion
The prior art made of record and not relied upon is considered to applicant’s disclosure.
Chiu et al US 20200082268 A1 (2020-03-12) (Abstract “An electronic apparatus and a compression method for an artificial neural network are provided. The compression method is adapted for the artificial neural network with a plurality of convolution layers. The compression method includes: setting a first pruning layer for coupling the first pruning layer to Lth convolution layer, where the first pruning layer has a plurality of first weighting values and each of the first weighting values corresponds to each of a plurality of channels of the Lth convolution layer; tuning the first weighting values, selecting a part of the channels of the Lth convolution layer to be at least one first redundancy channel according to the first weighting values, and generating a compressed Lth convolution layer by deleting the at least one first redundancy channel; and removing the first pruning layer, and generating a first compressed artificial neural network.”).
Rhodes et al US 20220051103 A1 (2022-02-17) (Abstract “An apparatus is provided to compress CNN models using a combination of filter pruning and tensor decomposition. For example, the apparatus accesses a trained CNN that includes convolutional tensors. The apparatus prunes the filters of a convolutional tensor to generate a sparse tensor.”).
Watson et al US 20210357817 A1 (2021-11-18) (Claims 1 “ pruning the reduced input MLM such that a first input of a plurality of inputs of the original MLM is not included as an input of the reduced input MLM; determining that a size of the reduced input MLM exceeds a memory threshold of a target device; and based on the determination that the size of the reduced input MLM is greater than the memory threshold of the target device, prune the reduced input MLM to remove a second input of the plurality of inputs of the original MLM included in the reduced input MLM to generate a second reduced MLM;”).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to URIAH V MOORE whose telephone number is (571)384-8341. The examiner can normally be reached Monday-Friday 8am-5pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jeanette Parker can be reached at (571) 270-3647. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/URIAH VENDELL MOORE/Examiner, Art Unit 2142 /Mariela Reyes/Supervisory Patent Examiner, Art Unit 2142
Prosecution Insights